CFUnited 2010 = Cache Me If You Can

July 28, 2010 2:45pm eastern

Dan Wilson, Mike Brunt, and Mike Allen

Dan Wilson:

shared resources = wait times

conversion rate fall off by landing page speed

http request/response model

Wikipedia site

Questions:

articles

revision

latest version

surges

ehcache - geodns

slide map of wikimedia

squid front cache layer

reverse proxy cache (front cache)

varnish/squid (free)

aicache (commercial)

reverse proxy placement

are very, very fast

used everywhere

caches whole response

not real time!!!!

small increment caching chart

5 seconds = 1/4 request per second

.5 seconds = 2 requests per second

no canche = 5000 requests per second

works for static content

wikipedia 60+ squid servers (cheap boxs lots of ram)

carp (cache array ... protocol)

in memory cache - key/value store

ehCache - distribute cache across multiple placements

memcached.get

memcached.set

What to cache?

Finite resources

goal is to keep load off server

strategies to cache

struct = querykey ‘ listingquery_make_#arguments.make#’

big struct in memory

argument for each filter item: problem is

how much data?

permutations

13.4 petabytes

two things hard in cs: caching and naming things

page generation process

weather site example

caching should be used more like a scaple and not a chainsaw

cf caching lives in the same jvm

Mike Allen:

key consideration when picking the right caching options

omdas? law - calculating bang for buck

right type of cache

ease of integration

feature sets